Artificial Intelligence: Comeback Chance for Japanese ManufacturingEconomy Science Technology
The Third AI Boom
AI is booming. Artificial intelligence has a history of about 60 years as a research field, dating back to 1956, when this name was adopted. The field experienced a boom first in the 1960s and again in the 1980s; the current AI boom, which started around 2010, is the third.
Some easy-to-see examples of AI at work have attracted popular notice, such as the development of a computer program that can beat professional human players at shōgi (Japanese chess), IBM’s Watson, which beat champions on a popular quiz program, and the Siri voice-interaction software on iPhones. But in technological terms the focus is now on what is called “deep learning,” a field in which research is progressing at a breakneck pace.
For example, it is hard for computers to recognize the objects in images—to identify a flower, for example, or a sailboat, or a coffee cup. And until recently it was thought that they would be unable to match humans at image-recognition tasks like these for decades to come. As Marvin Minsky, a leader in AI research, observed, the easier something is for a child, the tougher it is for a computer. Even something like playing with blocks was a challenge for computers. And image recognition was said to be a typical example of this sort of task—easy for humans but hard for computers.
Computers Surpass Humans in Image Recognition
But then, in 2012, a major advance was achieved in deep learning, and less than three years later computers had already surpassed humans in image recognition accuracy. This sort of ability was seen in programs developed by Microsoft and Google, announced in February and March this year, respectively. So computers are now better than humans at identifying objects and people in photographs. This is a huge step forward.
Deep learning, a form of representation learning, involves having computers learn what parts of the real world to focus on. In earlier AI approaches—and, one might well say, in every sort of engineering-based model—the unimportant elements were discarded and the key elements were turned into a model to allow efficient computation. And humans decided what parts of the real world computers would direct their vision toward. This was a big problem. Though there were various approaches designed to allow automatic computation, human involvement was indispensable at the initial stage. Deep learning has been making it possible to leave humans out of the loop. This is a highly significant development.
Rapid Progress in Corporations and Universities Overseas
The lead actors in this technological innovation are researchers mainly in the United States and Canada and corporations largely based in Silicon Valley. France is also catching up fast, tapping its strength in theoretical mathematics. And major Chinese firms are striving to get a piece of the action.
Google has for some time been devoting great efforts to AI research, and in 2013, the year after bursting on to the deep learning scene, it hired Geoffrey Hinton, a leading light in the field. Early in 2014 it followed up with the acquisition of DeepMind Technologies, a small British start-up, for about ¥40 billion. At the time many were surprised at the purchase, but now it looks like an excellent investment to have made.
Meanwhile, Facebook has set up AI research labs in New York, Menlo Park (Silicon Valley), and Paris, reportedly with tremendous budgets. The AI research effort is being lead by Professor Yann LeCun of New York University, who is French by birth. France has traditionally been strong in the field of theoretical mathematics, and it is emerging as a major player in AI now that this field of math has come to be a key component of deep learning. Facebook’s strategy seems to involve looking from the East Coast of the United States across the Atlantic to Europe.
Japan Lags Behind in the Second Pack
Baidu, China’s major search-engine company, has set up its own deep-learning research institute, headed by Andrew Ng, a star researcher from Stanford University. Ng is a Chinese American, educated in Hong Kong, Singapore, and the United States. China’s approach involves combining its big business capital with the talents of ethnic Chinese researchers working all around the United States.
Aside from the above three Internet giants, we see a proliferation of start-ups that are seeking to tap the potential of advanced AI and deep learning. The United States, which won decisively in the Internet race, is maintaining its overwhelming lead in the current competition to develop AI, the key technology for the years to come. Its closest rivals at this point are Asian: China’s Baidu and Tsinghua University, Hong Kong University, and the National University of Singapore. Japan is part of the second pack, which lags far behind the leaders.
The Japanese Pioneers of Deep Learning
Japan, however, has a well of strength not to be discounted. The original ideas that served as the basis for deep learning came from Japanese researchers. In 1980, Fukushima Kunihiko, a researcher at the NHK Science & Technical Research Laboratories, announced a pattern-recognition system that he dubbed “neocognitron.” Though it was good at recognizing written characters, it did not attract that much attention. It was only with the passage of time and the emergence of today’s computing technology that the true value of Fukushima’s system has become apparent.
Another key figure was Amari Shun’ichi, now a professor emeritus at the University of Tokyo, who did pioneering research on neural networks, providing the base for deep learning technology. Amari, who will turn 80 next year, is still full of vitality, and his name is often mentioned in presentations at academic conferences on deep learning.
Are these just additional examples of the often-observed phenomenon of individual Japanese playing key roles on the global scene? Actually there is more to the story. Consider the following passage:
“We will enhance white-collar productivity. For this purpose, we will make use not just of text but also of images and sounds.”
At a glance, this looks as if it could be the mission statement of a start-up, perhaps one of the current crop of new businesses aiming to make use of deep learning. It might also remind some readers of Google’s mission: “to organize the world‘s information and make it universally accessible and useful.”
Actually the quote is taken from the 1982 document proposing the launch of Japan’s Fifth-Generation Computer Systems project. Undertaken at a time before personal computers, this project received ¥57 billion in funding from the Ministry of International Trade and Industry (now the Ministry of Economy, Trade, and Industry), and it pursued research on AI. Numerous top-flight researchers visited Japan, and both the United States and European countries are said to have had serious discussions about strategies to respond to this Japanese initiative.
It seems that this 1982 enterprise was altogether too advanced for its time. The concept is one that would pass muster even today. Or perhaps it would be better to say that it finally passes muster today. People say that the project failed because it was too advanced and therefore ended up unable to find the right direction for development of technology. In those days there simply was not enough data. There was no Internet, no World Wide Web—and no way of raising white-collar productivity But when we look back at the project, I believe we can say that it was actually aiming in just the right direction.
One can imagine that if the web had emerged 10 years earlier, Japan might have taken the place that Silicon Valley now occupies. The Fifth-Generation Computer Systems project, conducted when Japan was bubbling with rapid growth, may be seen as marking the high point of our country’s approach to “number one” status. After learning about the existence of this project and doing some research on it, I am impressed not so much with its technological content as with the strong ambition and strategic direction underlying it—the quest to be number one.
Japan’s AI Potential
We can expect AI to continue to develop and to have a major impact on a broad range of sectors, such as manufacturing of products like automobiles and industrial machinery, social infrastructure in areas like transportation and logistics, security, including crime prevention, robotics, and medical and nursing care. The added value from this impact is likely to be tremendous.
AI is actually a field to which Japan is well suited. Let me cite a few reasons for this:1. Strong needs due to population shrinkage and aging
As Japan’s population declines, productivity must be raised, and there are strong social demands for AI and for the development of AI-equipped robots.2. Ample supply of AI-related human resources
Thanks to the Fifth-Generation Computer Systems project, Japan has ample human resources for AI. People who were university students at the time of the project are now serving as university professors and are educating the next crop of experts. The Japanese Society for Artificial Intelligence has 3,000 members, comparable in scale to the 5,000–6,000 members of the Advancement of Artificial Intelligence, the global AI organization. (By contrast, in most information-technology-related fields, the Japanese society membership figure is smaller by an order of magnitude.) Furthermore, Japan has many people who directly experienced the first and second AI booms, and the level of understanding of the field is high.3. Traits matching the requirements for AI
In the world of the Internet, value is sought through the linking of information, and the businesses that succeed are those that are quickest to latch on to popular needs. In the case of AI, though, the value system is shared, and what is required is the cleverness to fully understand the mathematical foundations, along with the diligence in fine-tuning the parameters. These are precisely the talents that manufacturing engineers have.4. No language barrier
In the case of the Internet, the language barrier between Japanese and English was a major impediment for Japan. But in the case of AI, what counts are the algorithms, and language ability is not a major consideration.5. Strong connection to hardware
With AI, there is a strong chance that Japan can apply its traditional winning pattern of incorporating technology into products and selling them. Deep learning ties in with sensor technology and robotics, and so Japan can probably tap its strengths in these fields.
Chance for a Manufacturing Revival
Considering how well suited Japan is to AI, it could provide a chance for our country’s manufacturing sector to shine again. The prospects should be especially good for companies in fields like sensors, security, logistics, robotics, and infrastructure. Their ability to use this chance may well hold the key to our country’s future.
So far we seem to be doing reasonably well within Japan. Both the government and businesses are keeping up with developments at a relatively rapid pace and working to take advantage of the opportunities that AI presents. On the business side, Dwango (a telecommunications and media company) has established an AI research lab, and a number of major corporations like Recruit are planning similar moves.
A wave of start-ups are also emerging. Preferred Infrastructure, the company with the greatest technological strength in the field, has established Preferred Networks as a firm specializing in deep learning and the “Internet of Things.” This is a star prospect for Japan’s manufacturing sector.
On the government side, in May this year METI established the Artificial Intelligence Research Center within the National Institute of Advanced Industrial Science and Technology. Tsuji Jun’ichi, who was previously at Microsoft’s research institute in China, has been placed in charge of the center, and the process of recruiting researchers and establishing the organizational framework is progressing steadily. The Ministry of Education, Culture, Sports, Science, and Technology, meanwhile is also conducting studies into AI as the next major current in science and technology. And the Ministry of Internal Affairs and Communications is looking ahead to the new prospects for communications and intelligence with deliberations that are even considering the “singularity” (when AI overtakes human thinking). For my own part, I have moved quickly to establish an AI teaching program at the University of Tokyo, which has received generous offers of support from corporations.
Aim to Lead the World in AI
Though we cannot predict the outcome, the chances for Japan are great. So far we are doing reasonably well. And we may be able to turn AI into a major legacy for the generations to come.
I hope we will be able to orient the AI boom in the right direction. AI offers the chance for a revival of Japanese manufacturing. It may allow us to maintain productivity and enjoy comfortable lives even as our country’s population contracts. Japan may be able to play the lead role in building the society of the future—a society of convenience, safety, and peace of mind, where people can live and work in a more human way.(Originally published in Japanese on August 10, 2015. Banner photo: Watson, IBM’s AI computer system, defeated a series of human opponents to win first place on the US quiz show Jeopardy! in 2011. The photo shows an exhibition round, with Watson positioned between two human contestants. © AP/Aflo.)